Agent Brain

Shared memory for your AI coding agents.

Self-hosted persistent memory, semantic caching, and cross-agent context for Claude, Codex, Gemini, Cursor, and Aider. Runs locally alongside your Docker stack. Nothing leaves your machine.

What it does

Persistent memory

Agents remember decisions, patterns, and context across sessions — stored locally in PostgreSQL + Qdrant.

Semantic cache

Similar queries hit the cache instead of burning tokens. 40–70% reduction measured on real workloads.

Cross-agent context

Claude Code and Codex working in the same repo share the same brain over MCP or REST.

Local-model brain

Ollama (qwen3, nomic-embed) handles extraction, scoring, and conflict resolution on your hardware.

Memory governance

Audit, deduplicate, and consolidate memories so retrieval stays focused as the corpus grows.

Desktop app

Tauri 2.x app with Docker lifecycle control, live dashboard, knowledge explorer, and cache monitor.

Install

Download the desktop app from Releases — macOS, Windows, and Linux builds. Or run from source:

git clone https://github.com/blackrock-engineering/agent-brain
cd agent-brain
cp .env.example .env
./scripts/setup.sh
docker compose -f docker-compose.yml -f docker-compose.dev.yml up -d

Connect your agents

MCP

Claude Code, Codex, Cursor, and Gemini CLI connect at http://localhost:3100/mcp.

REST + WebSocket

Anything else hits the brain engine at http://localhost:9090/api/v1.